Empath is a multi-agent NLP ML/AI “Conversational therapist/personal coach/VA”.
2018-current: Designed, coded, trained https://EmpathyBot.net, a natural language processing model initially using a home-brewed pattern-recognition system, pre-LLMs. Built interface for users to easily assist in training the model by critiquing its responses via multi-choice dropdowns for each part of the ML’s response. Next, created an API in Python/Flask to integrate NLTK to increase sentiment accuracy. Added Telegram bot. Added multi-model auto-switching between Claude 3 Opus, 3.5 Sonnet, and GPT-4o APIs, as needed. Four-agent prompt-chain functionality:
(a) “Compressor” reduces size of longer user queries to decrease cost.
(b) “Abuse Detector” determines how far user query is from bot’s intended knowledge domain and acts accordingly, which can include banning a user’s IP address.
(c) “Coach” processes user input and writes a draft response.
(d) “Editor” analyzes potential output for accuracy and kicks it back to “Coach” if needed.
– Currently: (a) Replacing last ASP code with Flask to remove dependence on MS IIS and (b) Incorporating Retrieval-Augmented Generation (RAG).
Full featured CMS I created in 1996 before WordPress was a dream. Includes API calls for live price and time data for UPS, USPS, FedEx, and UPS. Also includes API calls to PayPal and Authorize.net.
Relationships in the HotKey SQL database:
Python game: Zombie Feeder
Game, Source code, and entertaining video.
Python: Cellular automata simulation
Scott Swain created this simulation in alpha-beta behaviors in fish with Python. Frameworks used include Flask, Psycopg2, and Arcade for graphics. Below is a short video of the code in action with sound. Used Visual Studio Code as the IDE.
Built for PhdBookbinding using Classic ASP, VBscript, Javascript, MS Access, HTML, and CSS. Front-end drag & drop.